Autonomy vs the Auto Insurers — Digest
TL;DR
The insurance stack for autonomy is being built in public this cycle: Allianz signed a full fleet-and-product-liability program for Waymo's European rollout, AM Best declared US AV liability will land in surplus lines first, and Aurora's investor day sold out a 200-truck driverless order book with a 30,000-truck 2030 vision. Meanwhile Progressive posted its worst combined ratio of 2026 (89.3 in August) and the equity market is punishing the insurers — but a key check on the erosion thesis emerged: national VMT data shows no decline, suggesting Progressive's "4% VMT drop" is book mix, not the AV effect. The premium pool is still intact; the replacement infrastructure around it is not.
The liability map is being drawn by contract, not statute
The AV insurance question got its first institutional answers this cycle — and neither came from the personal-auto carriers.
"Motor fleet insurance, general and product liability programmes, and new insurance models designed for autonomous mobility platforms." — ZAG Daily (finding
)
That's the scope of the Waymo–Allianz Partners collaboration announced September 16, built for the European rollout starting with Germany — a global insurer constructing the entire claims-and-liability stack directly for the AV operator. AM Best's September 14 report explains why the US equivalent won't come from the admitted market: commercial auto has been "a consistently unprofitable P/C line of coverage for an extended period," so driverless coverage will "at least initially, be found in the surplus lines market" (Insurance Journal, finding). The severity math explains the reluctance: driver-assistance systems have cut claim frequency 13–39% while raising the cost of remaining claims, and a Waymo I-Pace carries roughly $100,000 in sensors that are "almost impossible to actually repair" per Root founder Alex Timm (TheStreet, finding
). The pool isn't shrinking yet — but the structures that will replace it are being standardized elsewhere.
What to watch: Whether a US admitted carrier lands a Waymo-style fleet mandate onshore — the first sign the pool's replacement economics are being built domestically (see the open thread below).
Waymo's fleet is compounding where the miles count
Waymo crossed from novelty to volume: 15 US cities, roughly 4,000 vehicles, and 500,000 paid robotaxi rides per week — with Texas as the growth engine.
"About 80% of Waymo's roughly 4,000 robotaxis are in California and Texas, and Texas is where the action is now: Waymo's fleet there has grown by nearly half in the past three weeks." — TechCrunch (finding
Texas registrations hit 1,102 AVs as of September 24, Austin freeways opened to rider-only trips (territory expanding from 37 to 150 square miles), and commercial service began in Denver, San Diego, and Tampa on September 1 (KXAN, finding). The safety dataset underwriting all of it keeps sharpening: 0.01 serious-injury crashes per million miles across 220 million rider-only miles versus a 0.23 human benchmark (TheStreet). Every one of those 500K weekly rides is a personally-insurable trip converted into fleet risk — the erosion happens one ride at a time, and the weekly run rate is now the number to track.
What to watch: Whether the Texas fleet growth rate (+49% in three weeks) holds as Ojai minivan deliveries scale toward MoffettNathanson's 5,100-import year-end estimate.
Trucking liability migration just got an order book
Aurora's investor day moved autonomous trucking from pilot to industrial scaling — and quantified who owns the risk.
"The Aurora Driver is engineered to be one of the safest and most experienced drivers on the road having safely completed over 500,000 driverless miles since commercial launch." — Aurora IR (finding
)
The company is fully allocated to exit 2026 with 200 driverless trucks under Transport-as-a-Service agreements, targeting more than 30,000 by 2030 with "SaaS-like gross margins" for its Driver-as-a-Service model. Aurora's trucks for McLane and Werner average over 225,000 annualized miles — more than double traditional utilization (Aurora IR). The DaaS wrinkle matters for insurers: Hirschbach's intent to own 500 trucks from 2027 means fleet customers will need their own programs — commercial-auto premium re-underwritten around software product liability, which per AM Best starts in surplus lines (finding). Uber-affiliated Neben Holdings block-sold 29.37 million Aurora shares on September 15 — an early backer trimming into the scaling story (finding
).
What to watch: Whether the first Hirschbach DaaS deliveries in 2027 come with a named insurance program — that contract will show where the liability actually sits.
Tesla's Cybercab is stress-testing self-certification
Tesla crossed into fare-charging robotaxi service and immediately collided with its regulator.
"NHTSA gave Tesla until September 30 to answer a new set of questions on whether it properly self-certified the Cybercab, including whether temporary human-driver controls or other equipment were part of the certification basis." — Startup Fortune, citing Reuters/Electrek (finding
)
The Cybercab — no wheel, no pedals — began charging fares in Austin on September 3, and within hours NHTSA opened an audit into whether the vehicle was ever legal to sell, with reported penalties reaching $139 million (Cognitive Market Research, finding). The analysis frames the industry fork neatly: Tesla's self-certification is "fast, uncapped, and risky," while Zoox's Part 555 exemption is "slower and capped at 2,500 vehicles a year, but it buys certainty." Meanwhile Tesla's insurance expansion stalled further — no Washington approval by the September 1 proposed effective date, and the Georgia application withdrawn earlier this year, the first known pullback from a new market (Drive Tesla Canada). The company is nonetheless soliciting interest in Cybercab fleet purchases, which would push its self-underwritten liability model into commercial fleet territory (finding
).
What to watch: The September 30 sworn-answer deadline — clearance legitimizes purpose-built AVs nationwide; rejection extends the product-liability timeline and forces a redesign.
The insurer scorecard turned, and the market noticed
Progressive's August monthly delivered the sector's sharpest underwriting deterioration of the year — and the stock tape finally moved with it.
"The largest single-month gap recorded so far this year, and the pattern that typically precedes a carrier tightening rate is taking shape." — Insurance Business (finding
)
Progressive's combined ratio hit 89.3 (up 6.2 points YoY), net income fell 22% to $951M, and operating EPS missed consensus — with July's deterioration occurring without comparable weather, suggesting the pressure predates August's storms (TradingKey). The market has started repricing: PGR down 8.7% over three months, ALL down 11.3% in a month, ROOT 48.8% below its high, KMPR 49.1% below — while TSLA is up 4.9% and AUR up 6.2% (finding). The asymmetry supports the mispricing thesis: AV disruption is being priced into platforms like Uber (down ~28% in a year, per TIKR), not into the $350B premium pool (TheStreet).
What to watch: Whether Progressive's September monthly confirms the underlying (non-weather) deterioration — two clean months would force a rate-tightening conversation.
What surprised us
- The national VMT check undercut the erosion headline. FHWA data via FRED shows moving 12-month VMT flat-to-slightly-up from February through July 2026 — no national 4% decline exists, pointing to Progressive's Q2 figure being book-specific mix rather than an AV effect (FRED, finding
- Berkshire is on both sides of the trade. GEICO's parent tripled its Alphabet stake in Q1 and added $10B via a June private placement — its Alphabet position is now Berkshire's third-largest holding (Business Insider, finding
- Root is the thesis in a straitjacket. Timm gave the sharpest articulation of the liability shift — "change the dynamics of who owns the liability, really change everything" (TheStreet) — with OEM partnerships (Toyota, Hyundai Capital, Carvana at 51% of new business) that look built for exactly this future. Yet revenue growth is stalled at +1.6% and the stock sits 48.8% below its high (finding
). The market is valuing the OEM-embedded future as if it never arrives — which is the mispricing thesis in miniature.
Open threads worth a vote
- Will a US admitted carrier land a Waymo-style AV fleet insurance mandate, or does US AV liability default to surplus lines and self-insurance? — AM Best says E&S first; Allianz just built the onshore-equivalent template in Europe. Your vote steers what we track next.